My Strategy Game Was Won 98% of the Time. Here Is How I Made It Matter.

On YOMP! The Road to Stanley, making a small game about a real war, and the day the numbers told me my decisions did not count.

YOMP! The Road to Stanley: free to play in your browser.

https://stronma.itch.io/yomp-the-road-to-stanley 

“Yomp” is Royal Marines slang for a long march carrying heavy kit, and in 1982 it became famous as the way British troops crossed East Falkland on foot to reach Stanley. Bog, wind, rock, weight on your back, and a long way still to go. It is the right word for this game, because more than anything, YOMP! is about how far you can push tired people before they cannot fight.

YOMP! The Road to Stanley v0.8 is now on itch.io. It is free, it runs in your browser, and it was built for a phone held upright, though it works on a desktop too.

What it is

You command a British brigade ashore on East Falkland. You land on 21 May 1982 and you have until 21 June to reach Stanley. To get there you need to secure at least three of the five mountain positions that guard the town (Mount Harriet, Two Sisters, Mount Longdon, Tumbledown and Wireless Ridge), and then move a formation into Stanley itself.

You have five units, each based on a real one and each with its own character: 2 PARA is aggressive, 3 PARA balanced, 40 Commando flexible, 42 Commando good on the mountains, and 45 Commando tires less on the march. Each turn is six hours, and you get a handful of Command Points to spend on moving, reconnoitring, calling in fire support, lifting troops by helicopter or simply resting. Then you press EXECUTE ORDERS and watch the clock run.

When you walk into an enemy position, there is a battle, and you make two decisions: how to go in (straight assault, recon and flank, or pull back), and then, on the battle net, whether to keep going, commit your reserve, call for naval gunfire or break contact. Along the way, events from the real campaign arrive: the loss of the Atlantic Conveyor and most of your helicopters, an SAS patrol report, a Falkland Islander with intelligence, boggy ground, trench foot.

Contact at Mount Harriet. How you go in matters far more in v0.8.

Making a game about a real war

Making a game about a real war needs care. The Falklands War was not long ago. Real people fought and died there, on both sides, and the Islanders lived through it. Plenty of people who were there are still with us. This should not be something flippant.

So the game follows a few rules. It starts from the real landings at San Carlos on 21 May, and from there it is honest about being a game: “History sets the board. You decide what happens next.” The other two landing options, at Cow Bay and Berkeley Sound, are labelled as fictional what-ifs. The map is schematic and not to scale. Some units are simply not in it: 5th Infantry Brigade, with the Guards and the Gurkhas, is missing, and I would rather say so than pretend otherwise. And there is a Historical Archive inside the game with the real dates for the landings, the Atlantic Conveyor, Goose Green and the mountain battles, so that anyone curious can tell the history from the game.

What I hope the game gets across, in its small way, is the weight of the problem: distance, ground, supply, exhaustion, and decisions made without knowing quite what is on the next hill.

How it was made

I am a hobby developer, not a studio. YOMP! is a single HTML file with its music built in (a 3 MB download), and I built it with a lot of help from an AI assistant, [CONFIRM: ChatGPT or Claude]. I would describe what I wanted, play what came back, and push back when it felt wrong. The ideas and the final call on every build were mine. [CONFIRM: If the v0.8 fixes and tuning were also done with an AI assistant, say so here.]

I designed it for phones first, in portrait, so a turn fits into a spare minute wherever you are. That brought its own problems, as the testing showed.

The uncomfortable discovery

Before releasing v0.8 I wanted to know whether the game was too hard. I am a terrible judge of my own game: I know where everything is and what every event does. So instead of guessing, the game’s real code was run thousands of times without graphics, with simulated commanders making every decision.

There were four of them. One clicked at random. One was a naive rusher that marched everyone straight at the nearest objective and never bothered with reconnaissance or fire support. One was a sensible commander that followed the sort of rules a careful newcomer would: advance together, reconnoitre the target, bring in fire support, rest when tired, flank rather than charge. And one was an expert that knew the hidden battle maths perfectly, which no human ever will. Every campaign was run on the same set of random seeds before and after my changes, so any difference would come from the changes, not from luck. A whole run of thousands of campaigns took about a minute.

The answer was clear, and surprising. v0.7 was not too hard. It was far too easy.

The sensible commander won 98% of campaigns on the standard difficulty. The naive rusher, who never prepared anything, won 93%. Even random clicking won 37%. Battles were won about 95 to 100% of the time, whatever the player chose. The brigade usually reached Stanley in two or three in-game days, so most of the campaign’s events, including the Atlantic Conveyor on 25 May, never even happened.

The part that really stung was a set of tests where the sensible commander was forced to drop one habit at a time: no fire support, no reconnaissance, always charge, random choices on the battle net. In v0.7, every single variant still won between 99.8% and 100%. None of the decisions I had built the game around made any difference.

Event choices are still flavour more than strategy. That is on the list for later.

Making the choices count

The fix was not one big change but a dozen small ones, all pointing the same way. The strength of the defenders now matters much more. Fatigue builds with the length of the march and the terrain, and tired troops fight badly, which is, after all, the point of a game called YOMP. Fire support is the main way to prepare an attack, but it costs supply, and the brigade now uses supply just by being in the field. A failed attack only dents the defenders instead of grinding them down. Argentine positions reinforce while you dither. Reconnaissance now feeds directly into how well a flanking attack goes. Resting away from the front recovers a little strength, so one bad battle does not have to start a death spiral.

And there are now three difficulty levels: RECRUIT for learning, COMMAND as the intended challenge, and VETERAN for experienced commanders who want a much stiffer test.

Here is what the simulated commanders do on v0.8 COMMAND. The sensible commander wins 41.8% of campaigns, or about 57% on the historical San Carlos landing. The expert still wins 99.5%, which tells me the game is beatable with good play. The naive rusher wins nothing at all, though on RECRUIT it gets through about one campaign in four. Random clicking wins nothing.

More importantly, the decisions matter now. Take away the sensible commander’s fire support and its win rate falls from 55% to 0.3% (on the San Carlos and Cow Bay landings). Take away its reconnaissance and it falls to 4%. Always pressing on in battle instead of committing the reserve or calling naval fire drops it to 13%. Goose Green, which I always wanted to be a real dilemma, finally is one: attacking it costs time and men, but bypassing it leaves a garrison raiding your supply line, and which is better depends on how well you fight.

The mountain belt around Stanley is where most campaigns are now won or lost.

Bugs found along the way

Running thousands of campaigns shakes out bugs that one person playing never finds. Ten turned up. The worst: if two of your formations attacked the same position in one turn, the second battle was fought even after the first had taken it, and losing it handed the position back. Reloading the page mid-battle let a formation stroll straight through an enemy position. You could queue more helicopter lifts than you had helicopters. On phones, the map circles were so small, and the unit markers overlapped so badly, that it was easy to cancel your own orders by accident, so there is now a formation list under the map. All fixed in v0.8.

What I still do not know

I want to be honest about the limits here. Every one of those numbers comes from simulated players. Bots are not people. Real players learn between campaigns, improvise, and make mistakes no rule-follower would make. Nobody has yet given me real-player numbers for v0.8, and it has only been tested in emulated phone and desktop browsers, not on a real iPhone or in Safari.

There are also things I know are still thin. Weather is purely cosmetic. Morale hardly moves. Helicopter lifts make no measurable difference, events arrive on a fixed schedule, and event choices change results by only a few points. And the game is short: a COMMAND win takes three to five in-game days. Those are problems for future versions, and I would rather tell you about them than have you discover them.

Come and play

YOMP! The Road to Stanley v0.8 is free to play in your browser on itch.io: [itch.io link]

Start on RECRUIT for your first campaign. Reconnoitre before you attack, bring in fire support, and rest your troops before the mountains. And then, whether you reach Stanley or end up stuck in the bog with your supply gone, please leave a comment on the itch.io page and tell me how it went. You will be the first real players, and that is the test that counts.

— Emperormarkus

Scroll to Top